DocumentationShowcasesBack to top
Skip to content
Showcases
Rebyte Docs
Agents API
OverviewQuickstartShowcasesRebyte Agent SDKUpgrade to 0.3.0OpenAI compatibilityArchitectureModelsConfiguring Agents
Sessions
Run and continue sessionsEvents and conversation historyManage sessionsWebhooks
Environments and sandboxes
Rebyte SandboxesSandbox lifecycleSandbox securityFiles and artifacts
Tools and integrations
Dynamic WorkflowDecision toolWeb searchFunctionsMCP connectionsPlatform connectionsPluginsVaults
Workflow AgentsScheduled AgentsMulti-agentObservability and usageTracing

Not yet supported

Developer documentation/Agents API

Showcases

Explore Commerce Agent and Impo, two open-source applications built with the Rebyte Agents API.

Agents API4 min read
On this pageCommerce AgentHow it uses the APIExplore the implementationImpoHow it uses the APIExplore the implementationStart your own application

These projects show how to build your own product around managed Agent execution. For a tour of both applications, read Two apps built with the Rebyte Agents API.

ApplicationUser experienceAPI patterns to explore
Commerce AgentProduct discovery, comparisons, and a shopping cart inside a storefront.Sessions, deferred function tools, Skills, and streamed presentation.
ImpoA personal agent with chat, delegated tasks, and briefings.Saved Agents, independent Sessions, event streaming, and application-owned tool dispatch.

Commerce Agent

Commerce Agent adapts Anthropic's ACME commerce demo to the Rebyte Agents API. A shopper can ask for products, compare options, add items to a cart, and see a checkout summary. The storefront renders product cards and other UI from validated presentation tools.

How it uses the API

  1. Create a saved Agent from the included configuration, with twenty commerce functions, Web Search, and tool_search. The commerce functions use deferred loading.
  2. Create a separate Session for each browser conversation, with a hosted Environment and five inline Skill packages. Reuse that Session on later turns.
  3. Subscribe to Session events, then submit the shopper's message.
  4. When the runtime requests a function handoff, the Python application validates and executes the catalog, cart, or presentation function and returns its result.
  5. Stream text and presentation updates into the storefront.

Rebyte manages Agent execution and the Session environment. The application owns catalog data, shopping rules, cart state, and UI rendering. Deferred loading lets the model discover the function definitions it needs during a conversation. The Sandbox is allocated when an environment tool is used; a catalog function call alone does not require one.

ACME and its catalog are fictional. Checkout produces a summary without placing an order or charging a card. This Rebyte integration covers the retail shopping path; the repository's other verticals and merchant runtimes have separate setup.

Explore the implementation

See Function tools and Hosted environments and Skills to adapt these patterns to your own application.

Impo

Impo is an open-source personal agent with native iOS and Android clients. Users can chat, delegate tasks, capture spoken context with Echo, and receive personal briefings through Brief. Its shared backend integrates with Rebyte; a web client is planned.

How it uses the API

  1. The application authenticates the user and records accepted work in its database.
  2. A server worker creates or reuses a saved Agent for the user's main conversation. Delegated tasks use independent Sessions with inline Agent configurations.
  3. The worker subscribes to Session events and reads Turns and Items to reconcile execution with the application's conversation history.
  4. Function calls go through Impo's dispatcher. External-service tools run on the server; permitted device tools run on the selected native client.
  5. Clients receive streamed updates. A disconnected client can recover history; ending a stream subscription does not cancel the Agent's work.

Impo owns authentication, user memory, device permissions, tool dispatch, and background coordination. Rebyte owns managed Agent execution and runtime state. Temporal schedules Impo's Brief work, which invokes a Rebyte Agent; this is an application scheduling flow, not an integration with Rebyte's Schedule API. Echo transcription uses a separate Gemini pipeline, and its evidence feeds Impo's memory system.

Explore the implementation

  • Explore Impo: product overview and current app availability.
  • Source and setup: native clients, shared backend, and feature status.
  • Architecture: ownership, workers, storage, and device capabilities.
  • Rebyte gateway: Agent and Session calls, event subscriptions, and tool results.

See Sessions, Events and streaming, and Function tools for the underlying API contracts.

Start your own application

Follow the Quickstart for your first Session, or use the Rebyte Agent SDK for React hooks, a chat UI, and a server adapter. Start from the Commerce repository for a shopping experience; study Impo for a native app with durable background work and device tools.

Protocol referenceOpenAI Agents API ↗